Skip to main content

Table 1 SVM classification error rates on the test set of Data-I

From: Top scoring pairs for feature selection in machine learning and applications to cancer outcome prediction

  ρ= 0 ρ= 0.45 ρ= 0.6
Level TSP Fisher RFE TSP Fisher RFE TSP Fisher RFE
1 40.4 39.0 40.2 35.0 39.5 34.4 31.4 39.9 33.2
2 36.6 33.0 38.0 29.9 34.6 29.4 24.4 38.8 26.2
3 33.6 31.1 35.8 26.4 32.6 26.8 20.6 36.8 25.6
4 32.0 29.4 34.6 23.4 31.3 26.0 18.4 35.7 22.0
5 31.9 28.6 35.0 20.8 30.7 24.8 15.4 36.0 22.5
10 30.2 27.6 31.8 16.5 27.7 24.2 10.4 32.1 18.2
20 27.4 25.9 30.0 15.6 24.4 20.1 8.0 27.2 15.6
30 27.4 26.0 28.3 15.3 23.5 21.0 8.2 24.1 16.8
40 26.8 23.7 26.6 15.4 22.4 20.8 9.8 23.1 17.7
50 25.8 24.0 25.8 16.2 22.6 19.6 11.7 22.7 18.2
60 26.0 25.0 25.2 16.6 22.1 19.2 12.4 21.7 17.7
70 25.0 24.2 24.3 16.8 21.5 19.7 13.7 22.0 16.7
80 24.9 24.6 24.8 16.3 21.1 20.2 14.3 21.3 17.4
90 25.6 24.3 24.2 17.4 21.0 20.2 15.0 21.7 16.8
100 25.1 24.9 24.1 18.5 21.4 20.2 15.9 20.8 17.3
All features 24.1 24.1 24.1 21.4 21.4 21.4 20.6 20.6 20.6
  1. The error rates (%, mean) are shown at various selection levels as correlation varies among signal genes, using TSP, Fisher and RFE as feature selection methods